ª1 > a2 > a3, son las tensiones principales en el punto considerado.
4.4. FABRICACIÓN DEL TANQUE ESPESADOR DE LÁMINAS 1 Inspección del material.
4.4.5. Soldadura de las estructuras armadas.
4.4.5.1. Descripción del Proceso.
Discrete choice models have been developed using the data from the choice experiments (see Box 2.2 for the theoretical background on discrete choice modelling). The models developed from the SPDCE data are logit models, with two choice alternatives, described by attributes and levels as illustrated in Figure 3.4–Figure 3.6.
The estimation procedure assumes that respondents choose the alternatives with the highest utility. The outputs from the estimation procedure are attribute coefficients that best represent the choices made by the respondents. Both the values of the coefficients (in utility terms) and the significance of the coefficients are calculated and reported.
The ratio of coefficients quantifies the marginal rate of substitution between the attributes – the trade-off rate between one attribute and another. The ratios of the service coefficients and the cost coefficient provide an estimation of consumers’ WTP for service attributes, measured as the WTP additional money for increased stamp or parcel prices. WTP values and their significance are calculated and reported.
- Separate models were developed for businesses and residents, for each choice experiment.
- Likelihood ratio tests were undertaken for businesses to examine whether differences in preferences were observed between respondents in the different member states and between SMEs and large businesses; we found that merging the data across countries did not reduce the fit of the model across the choices significantly, so the pooling of data across countries was justified, but retaining separate models for SMEs and large businesses significantly improved the fit of the model and therefore separate models have been developed for SMEs and large businesses.
- Likelihood ratio tests were undertaken for residents to examine whether significant differences in preferences were observed between respondents in the different member states and between vulnerable and non-vulnerable residents; we found that retaining separate models across countries significantly improved the fit of the model, but merging vulnerable and non-vulnerable respondents did not reduce the model fit significantly; thus we pooled the data for vulnerable and non-vulnerable residents, retaining significant differences where warranted, but retained separate models for the different member states.
In pooling the data across countries for businesses, we converted the costs presented in the exercise into a common standard using purchase power standards (using the latest adjustments published for 2010).20 The models are set up as multinomial logit models,
pooling data across the countries but allowing for different error variance across countries, where the error variance for responses from Poland and Italy are measured relative to Sweden (Bradley and Daly, 1991).
For residents, we pooled the vulnerable and non-vulnerable responses for each country, assuming a multinomial logit model structure but allowing for different error variance between these groups (the error variance for vulnerable responses is measured relative to that for non-vulnerable responses). We have also translated the costs into one comparable unit using purchase power standards, so that we can compare findings across countries. For resident and business models, in cases where two cost levels were presented for two- class service options we used the average cost to represent the cost of the option.
For businesses and residents we also present the final findings in the local currency of the member state.
For business and resident models we have excluded respondents who the interviewer judged did not understand the SP questions or who did not give the questions much consideration (less than 5% of the data for business and 5.1% for Italian residents, 6.6% for Polish residents and 0.6% for Swedish residents).
A key part of the model analysis was to investigate how preferences of the sample regarding the importance of different postal attributes vary as a result of socio-economic characteristics, business characteristics, mail usage and internet provision of respondents. The characteristics that were examined in this investigation included:
RAND Europe Postal service preferences for customers in
Accent Italy, Poland and Sweden
Swiss Economics
for residents:
- gender
- age
- presence of longstanding heath problems - annual household income
- vulnerable or not
- access to internet or broadband access at home or elsewhere - use of internet to buy products.
for businesses:
- country
- number of employees
- annual turnover of company - sales by internet and deliver by post
- description of the organisation’s use of the internet - industrial sector.
For both residents and businesses: - volume of letters sent
- volume of letters received - volume of parcels sent - volume of parcels received
- letters or parcels sent using other services than the country's main provider - type of mail sent
- frequency visiting the post office
- access to internet or broadband access at home or elsewhere - location: city centre, rural or urban area.
Tests were undertaken comparing the predicted probabilities of choosing alternatives against the observed frequencies of choices, across each of these different respondent characteristics. Where these tests indicated significant differences in the value of attributes, the model specification was developed to take explicit account of this difference. In general, we have found little significant variation in preferences for the postal service attributes tested in this study across these dimensions. These findings are discussed in detail in the following sections.
We have also examined variations in cost sensitivity by income and vulnerability status for residents. Again, we found little significant variation in cost sensitivity across these dimensions and the results are discussed in the following sections.
During the development of the models the repeated nature of the data was not taken into account; it was assumed that each observation was independent, even though each respondent provided multiple responses. This assumption is incorrect as each respondent participated in three SPDCEs and provided multiple choice observations in each. Naïve models that do not take account of the repeated choice nature in SP datasets underestimate the standard errors on the coefficient estimates finding higher levels of statistical significance than would be judged once the repeated measured property of the data is taken
into account. Therefore, as a final step in the estimation procedure, a bootstrap re- sampling procedure was applied to the models to correct for model misspecification and take into account the repeated nature of the SP data. This procedure ensures that the t- ratios produced by the models are a realistic statement of the true errors of the model parameters.
The best final bootstrapped models are presented in Appendix B.